virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2025-08-04 22:03:09 +02:00
Code Issues Releases Activity

vulkan_decode: support multiple image views

Browse Source 
Enables non-monochrome video decoding using all our existing functions
in the context of an SDR decoder.
This commit is contained in:
Lynne
2025-02-21 01:33:54 +00:00
parent 491b65e343
commit 4495802bdb
5 changed files with 60 additions and 54 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
libavcodec/vulkan_av1.c
View File

@ -123,7 +123,7 @@ static int vk_av1_fill_pict(AVCodecContext *avctx, const AV1Frame **ref_src,
.codedExtent = (VkExtent2D){ pic->f->width, pic->f->height }, .codedExtent = (VkExtent2D){ pic->f->width, pic->f->height },
.baseArrayLayer = ((has_grain || dec->dedicated_dpb) && ctx->common.layered_dpb) ? .baseArrayLayer = ((has_grain || dec->dedicated_dpb) && ctx->common.layered_dpb) ?
hp->frame_id : 0, hp->frame_id : 0,
.imageViewBinding = vkpic->img_view_ref, .imageViewBinding = vkpic->view.ref[0],
}; };
*ref_slot = (VkVideoReferenceSlotInfoKHR) { *ref_slot = (VkVideoReferenceSlotInfoKHR) {
@ -346,7 +346,7 @@ static int vk_av1_start_frame(AVCodecContext *avctx,
.codedOffset = (VkOffset2D){ 0, 0 }, .codedOffset = (VkOffset2D){ 0, 0 },
.codedExtent = (VkExtent2D){ pic->f->width, pic->f->height }, .codedExtent = (VkExtent2D){ pic->f->width, pic->f->height },
.baseArrayLayer = 0, .baseArrayLayer = 0,
.imageViewBinding = vp->img_view_out, .imageViewBinding = vp->view.out[0],
}, },
}; };

90
libavcodec/vulkan_decode.c
View File

@ -130,9 +130,11 @@ static void init_frame(FFVulkanDecodeContext *dec, FFVulkanDecodePicture *vkpic)
FFVulkanFunctions *vk = &ctx->s.vkfn; FFVulkanFunctions *vk = &ctx->s.vkfn;
vkpic->dpb_frame = NULL; vkpic->dpb_frame = NULL;
vkpic->img_view_ref = VK_NULL_HANDLE; for (int i = 0; i < AV_NUM_DATA_POINTERS; i++) {
vkpic->img_view_out = VK_NULL_HANDLE; vkpic->view.ref[i] = VK_NULL_HANDLE;
vkpic->img_view_dest = VK_NULL_HANDLE; vkpic->view.out[i] = VK_NULL_HANDLE;
vkpic->view.dst[i] = VK_NULL_HANDLE;
}
vkpic->destroy_image_view = vk->DestroyImageView; vkpic->destroy_image_view = vk->DestroyImageView;
vkpic->wait_semaphores = vk->WaitSemaphores; vkpic->wait_semaphores = vk->WaitSemaphores;
@ -149,14 +151,14 @@ int ff_vk_decode_prepare_frame(FFVulkanDecodeContext *dec, AVFrame *pic,
/* If the decoder made a blank frame to make up for a missing ref, or the /* If the decoder made a blank frame to make up for a missing ref, or the
* frame is the current frame so it's missing one, create a re-representation */ * frame is the current frame so it's missing one, create a re-representation */
if (vkpic->img_view_ref) if (vkpic->view.ref[0])
return 0; return 0;
init_frame(dec, vkpic); init_frame(dec, vkpic);
if (ctx->common.layered_dpb && alloc_dpb) { if (ctx->common.layered_dpb && alloc_dpb) {
vkpic->img_view_ref = ctx->common.layered_view; vkpic->view.ref[0] = ctx->common.layered_view;
vkpic->img_aspect_ref = ctx->common.layered_aspect; vkpic->view.aspect_ref[0] = ctx->common.layered_aspect;
} else if (alloc_dpb) { } else if (alloc_dpb) {
AVHWFramesContext *dpb_frames = (AVHWFramesContext *)ctx->common.dpb_hwfc_ref->data; AVHWFramesContext *dpb_frames = (AVHWFramesContext *)ctx->common.dpb_hwfc_ref->data;
AVVulkanFramesContext *dpb_hwfc = dpb_frames->hwctx; AVVulkanFramesContext *dpb_hwfc = dpb_frames->hwctx;
@ -166,13 +168,13 @@ int ff_vk_decode_prepare_frame(FFVulkanDecodeContext *dec, AVFrame *pic,
return AVERROR(ENOMEM); return AVERROR(ENOMEM);
err = ff_vk_create_view(&ctx->s, &ctx->common, err = ff_vk_create_view(&ctx->s, &ctx->common,
&vkpic->img_view_ref, &vkpic->img_aspect_ref, &vkpic->view.ref[0], &vkpic->view.aspect_ref[0],
(AVVkFrame *)vkpic->dpb_frame->data[0], (AVVkFrame *)vkpic->dpb_frame->data[0],
dpb_hwfc->format[0], !is_current); dpb_hwfc->format[0], !is_current);
if (err < 0) if (err < 0)
return err; return err;
vkpic->img_view_dest = vkpic->img_view_ref; vkpic->view.dst[0] = vkpic->view.ref[0];
} }
if (!alloc_dpb || is_current) { if (!alloc_dpb || is_current) {
@ -180,15 +182,15 @@ int ff_vk_decode_prepare_frame(FFVulkanDecodeContext *dec, AVFrame *pic,
AVVulkanFramesContext *hwfc = frames->hwctx; AVVulkanFramesContext *hwfc = frames->hwctx;
err = ff_vk_create_view(&ctx->s, &ctx->common, err = ff_vk_create_view(&ctx->s, &ctx->common,
&vkpic->img_view_out, &vkpic->img_aspect, &vkpic->view.out[0], &vkpic->view.aspect[0],
(AVVkFrame *)pic->data[0], (AVVkFrame *)pic->data[0],
hwfc->format[0], !is_current); hwfc->format[0], !is_current);
if (err < 0) if (err < 0)
return err; return err;
if (!alloc_dpb) { if (!alloc_dpb) {
vkpic->img_view_ref = vkpic->img_view_out; vkpic->view.ref[0] = vkpic->view.out[0];
vkpic->img_aspect_ref = vkpic->img_aspect; vkpic->view.aspect_ref[0] = vkpic->view.aspect[0];
} }
} }
@ -201,41 +203,41 @@ int ff_vk_decode_prepare_frame_sdr(FFVulkanDecodeContext *dec, AVFrame *pic,
{ {
int err; int err;
FFVulkanDecodeShared *ctx = dec->shared_ctx; FFVulkanDecodeShared *ctx = dec->shared_ctx;
AVHWFramesContext *frames = (AVHWFramesContext *)pic->hw_frames_ctx->data;
vkpic->slices_size = 0; vkpic->slices_size = 0;
if (vkpic->img_view_ref) if (vkpic->view.ref[0])
return 0; return 0;
init_frame(dec, vkpic); init_frame(dec, vkpic);
if (ctx->common.layered_dpb && alloc_dpb) { for (int i = 0; i < av_pix_fmt_count_planes(frames->sw_format); i++) {
vkpic->img_view_ref = ctx->common.layered_view; if (alloc_dpb) {
vkpic->img_aspect_ref = ctx->common.layered_aspect; vkpic->dpb_frame = vk_get_dpb_pool(ctx);
} else if (alloc_dpb) { if (!vkpic->dpb_frame)
vkpic->dpb_frame = vk_get_dpb_pool(ctx); return AVERROR(ENOMEM);
if (!vkpic->dpb_frame)
return AVERROR(ENOMEM);
err = ff_vk_create_imageview(&ctx->s, err = ff_vk_create_imageview(&ctx->s,
&vkpic->img_view_ref, &vkpic->img_aspect_ref, &vkpic->view.ref[i], &vkpic->view.aspect_ref[i],
vkpic->dpb_frame, 0, rep_fmt); vkpic->dpb_frame, i, rep_fmt);
if (err < 0) if (err < 0)
return err; return err;
vkpic->img_view_dest = vkpic->img_view_ref; vkpic->view.dst[i] = vkpic->view.ref[i];
} }
if (!alloc_dpb || is_current) { if (!alloc_dpb || is_current) {
err = ff_vk_create_imageview(&ctx->s, err = ff_vk_create_imageview(&ctx->s,
&vkpic->img_view_out, &vkpic->img_aspect, &vkpic->view.out[i], &vkpic->view.aspect[i],
pic, 0, rep_fmt); pic, i, rep_fmt);
if (err < 0) if (err < 0)
return err; return err;
if (!alloc_dpb) { if (!alloc_dpb) {
vkpic->img_view_ref = vkpic->img_view_out; vkpic->view.ref[i] = vkpic->view.out[i];
vkpic->img_aspect_ref = vkpic->img_aspect; vkpic->view.aspect_ref[i] = vkpic->view.aspect[i];
}
} }
} }
@ -467,7 +469,7 @@ int ff_vk_decode_frame(AVCodecContext *avctx,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = vkf->img[0], .image = vkf->img[0],
.subresourceRange = (VkImageSubresourceRange) { .subresourceRange = (VkImageSubresourceRange) {
.aspectMask = vp->img_aspect, .aspectMask = vp->view.aspect[0],
.layerCount = 1, .layerCount = 1,
.levelCount = 1, .levelCount = 1,
}, },
@ -523,7 +525,7 @@ int ff_vk_decode_frame(AVCodecContext *avctx,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = rvkf->img[0], .image = rvkf->img[0],
.subresourceRange = (VkImageSubresourceRange) { .subresourceRange = (VkImageSubresourceRange) {
.aspectMask = rvp->img_aspect_ref, .aspectMask = rvp->view.aspect_ref[0],
.layerCount = 1, .layerCount = 1,
.levelCount = 1, .levelCount = 1,
}, },
@ -533,7 +535,7 @@ int ff_vk_decode_frame(AVCodecContext *avctx,
} }
} }
} else if (vp->decode_info.referenceSlotCount || } else if (vp->decode_info.referenceSlotCount ||
vp->img_view_out != vp->img_view_ref) { vp->view.out[0] != vp->view.ref[0]) {
/* Single barrier for a single layered ref */ /* Single barrier for a single layered ref */
err = ff_vk_exec_add_dep_frame(&ctx->s, exec, ctx->common.layered_frame, err = ff_vk_exec_add_dep_frame(&ctx->s, exec, ctx->common.layered_frame,
VK_PIPELINE_STAGE_2_VIDEO_DECODE_BIT_KHR, VK_PIPELINE_STAGE_2_VIDEO_DECODE_BIT_KHR,
@ -580,12 +582,14 @@ void ff_vk_decode_free_frame(AVHWDeviceContext *dev_ctx, FFVulkanDecodePicture *
av_buffer_unref(&vp->slices_buf); av_buffer_unref(&vp->slices_buf);
/* Destroy image view (out) */ /* Destroy image view (out) */
if (vp->img_view_out && vp->img_view_out != vp->img_view_dest) for (int i = 0; i < AV_NUM_DATA_POINTERS; i++) {
vp->destroy_image_view(hwctx->act_dev, vp->img_view_out, hwctx->alloc); if (vp->view.out[i] && vp->view.out[i] != vp->view.dst[i])
vp->destroy_image_view(hwctx->act_dev, vp->view.out[i], hwctx->alloc);
/* Destroy image view (ref, unlayered) */ /* Destroy image view (ref, unlayered) */
if (vp->img_view_dest) if (vp->view.dst[i])
vp->destroy_image_view(hwctx->act_dev, vp->img_view_dest, hwctx->alloc); vp->destroy_image_view(hwctx->act_dev, vp->view.dst[i], hwctx->alloc);
}
av_frame_free(&vp->dpb_frame); av_frame_free(&vp->dpb_frame);
} }

12
libavcodec/vulkan_decode.h
View File

@ -85,11 +85,13 @@ typedef struct FFVulkanDecodeContext {
typedef struct FFVulkanDecodePicture { typedef struct FFVulkanDecodePicture {
AVFrame *dpb_frame; /* Only used for out-of-place decoding. */ AVFrame *dpb_frame; /* Only used for out-of-place decoding. */
VkImageView img_view_ref; /* Image representation view (reference) */ struct {
VkImageView img_view_out; /* Image representation view (output-only) */ VkImageView ref[AV_NUM_DATA_POINTERS]; /* Image representation view (reference) */
VkImageView img_view_dest; /* Set to img_view_out if no layered refs are used */ VkImageView out[AV_NUM_DATA_POINTERS]; /* Image representation view (output-only) */
VkImageAspectFlags img_aspect; /* Image plane mask bits */ VkImageView dst[AV_NUM_DATA_POINTERS]; /* Set to img_view_out if no layered refs are used */
VkImageAspectFlags img_aspect_ref; /* Only used for out-of-place decoding */ VkImageAspectFlags aspect[AV_NUM_DATA_POINTERS]; /* Image plane mask bits */
VkImageAspectFlags aspect_ref[AV_NUM_DATA_POINTERS]; /* Only used for out-of-place decoding */
} view;
VkSemaphore sem; VkSemaphore sem;
uint64_t sem_value; uint64_t sem_value;

4
libavcodec/vulkan_h264.c
View File

@ -98,7 +98,7 @@ static int vk_h264_fill_pict(AVCodecContext *avctx, H264Picture **ref_src,
.codedOffset = (VkOffset2D){ 0, 0 }, .codedOffset = (VkOffset2D){ 0, 0 },
.codedExtent = (VkExtent2D){ pic->f->width, pic->f->height }, .codedExtent = (VkExtent2D){ pic->f->width, pic->f->height },
.baseArrayLayer = ctx->common.layered_dpb ? dpb_slot_index : 0, .baseArrayLayer = ctx->common.layered_dpb ? dpb_slot_index : 0,
.imageViewBinding = vkpic->img_view_ref, .imageViewBinding = vkpic->view.ref[0],
}; };
*ref_slot = (VkVideoReferenceSlotInfoKHR) { *ref_slot = (VkVideoReferenceSlotInfoKHR) {
@ -471,7 +471,7 @@ static int vk_h264_start_frame(AVCodecContext *avctx,
.codedOffset = (VkOffset2D){ 0, 0 }, .codedOffset = (VkOffset2D){ 0, 0 },
.codedExtent = (VkExtent2D){ pic->f->width, pic->f->height }, .codedExtent = (VkExtent2D){ pic->f->width, pic->f->height },
.baseArrayLayer = 0, .baseArrayLayer = 0,
.imageViewBinding = vp->img_view_out, .imageViewBinding = vp->view.out[0],
}, },
}; };

4
libavcodec/vulkan_hevc.c
View File

@ -164,7 +164,7 @@ static int vk_hevc_fill_pict(AVCodecContext *avctx, HEVCFrame **ref_src,
.codedOffset = (VkOffset2D){ 0, 0 }, .codedOffset = (VkOffset2D){ 0, 0 },
.codedExtent = (VkExtent2D){ pic->f->width, pic->f->height }, .codedExtent = (VkExtent2D){ pic->f->width, pic->f->height },
.baseArrayLayer = ctx->common.layered_dpb ? pic_id : 0, .baseArrayLayer = ctx->common.layered_dpb ? pic_id : 0,
.imageViewBinding = vkpic->img_view_ref, .imageViewBinding = vkpic->view.ref[0],
}; };
*ref_slot = (VkVideoReferenceSlotInfoKHR) { *ref_slot = (VkVideoReferenceSlotInfoKHR) {
@ -823,7 +823,7 @@ static int vk_hevc_start_frame(AVCodecContext *avctx,
.codedOffset = (VkOffset2D){ 0, 0 }, .codedOffset = (VkOffset2D){ 0, 0 },
.codedExtent = (VkExtent2D){ pic->f->width, pic->f->height }, .codedExtent = (VkExtent2D){ pic->f->width, pic->f->height },
.baseArrayLayer = 0, .baseArrayLayer = 0,
.imageViewBinding = vp->img_view_out, .imageViewBinding = vp->view.out[0],
}, },
}; };